Background: Performing axillary lymph node dissection (ALND) is the current standard option after a positive sentinel lymph node (SLN). However, whether 1-2 metastatic SLNs require ALND is debatable. The probability of metastasis in non-sentinel lymph nodes (NSLNs) can be calculated using nomograms. In this study, we developed an individualized model using machine-learning (ML) methods to select potential variables, which influence NSLN metastasis.
Materials and methods: Cohorts of patients with early breast cancer who underwent SLN biopsy and ALND between 2012 and 2021 were created (training cohort, N 157 and validation cohort, N 58) for the development of the nomogram. Three ML methods were trained in the training set to create a strong predictive model. Finally, the multiple iterations of the least absolute shrinkage and selection operator regression method were used to determine the variables associated with NSLN status.
Results: Four independent variables (positive SLN number, absence of lymph node hilum, lymphovascular invasion (LVI), and total number of SLNs harvested) were combined to generate the nomogram. The area under the receiver operating characteristic curve (AUC) value of 0.759 was obtained in the entire set. The AUC values for the training set and the test set were 0.782 and 0.705, respectively. The Hosmer-Lemeshow test of the model fit accuracy was identified with p = 0.759.
Conclusion: This study developed a nomogram that incorporates ultrasound (US)-related variables using the ML method and serves to clinically predict the non-metastatic status of NSLN and help in the selection of the appropriate treatment option.
Keywords: breast neoplasms; machine learning; nomogram; sentinel lymph node; ultrasound.
Copyright © 2022 Wu, Deng, Jiang and Zhang.